Brussels Mirage: The EU AI Act’s Subtle Shine across International Borders

While the U.S. is discussing a 10-year moratorium on all federal AI legislation (‘One Big Beautiful Bill Act’), many countries around the world are – in very different ways – legislating on AI (see global map). In the absence of a U.S. regulatory model, do countries with their own AI Bills (Japan) and AI Bill proposals (Brazil) look at the 2024 EU AI Act as a framework for guidance? This question gets to the heart of current debates surrounding the regulation of technology: Is there (or will there be) a Brussels effect of the EU AI Act?             

In drafting and adopting the AI Act in 2024, the European legislator sought – as a first move – to follow the regulatory course set by the 2012 General Data Protection Regulation (GDPR), while also advancing a distinct approach from that of the United States. Meanwhile, however, the EU is increasingly under pressure to justify its legislative interventions in the tech sector. Concerns over competitiveness (Mario Draghi, ‘The Future of European Competitiveness’) have contributed to what is described as a ‘de‑regulatory turn’: The European Commission has announced its intention to withdraw the proposed AI Liability Directive. At the same time, under the already enacted AI Act, it has drafted a Code of Practice for General Purpose AI, adopting a light-touch approach that has raised concerns among various stakeholders. The second Trump presidency, amid an unsettled geopolitical and geoeconomic context, adds to the strain on a European economy already trailing in the domain of artificial intelligence.

In this shifting landscape, the EU AI Act casts a mirage – projecting its regulatory model onto other jurisdictions, which are, however, increasingly placing their own distinct emphasis on AI regulation. The EU AI Act’s influence beyond Europe vanishes in light of the differing legal and cultural values that shape AI regulation. I call this the Brussels mirage.

Brussels effect of the EU AI Act?

The term ‘Brussels effect’, coined by legal scholar Anu Bradford in 2012, refers to the EU’s capacity to shape global regulation by extending the reach of its own legislative decisions. From the GDPR to the Digital Markets Act, the Digital Services Act, and now the AI Act, the European policymakers’ aspiration is that European law will de facto become the baseline for companies – no matter where they operate. Beyond that, governments and parliaments worldwide are expected to de iure draw inspiration from European legislation. In data protection, the Brussels effect is most striking, as seen in Brazil’s GDPR‑inspired data protection law.

That said, I contend AI regulation has changed the game: The EU AI Act does not generate a Brussels effect, as it is closely – though not overtly – tied to international AI standards predominantly influenced by non-EU actors. In terms of legally integrating AI standards, even countries like Brazil, which once leaned towards EU regulation, are increasingly adopting their own approach to AI governance. The EU AI Act’s influence is thus impressive from afar, but elusive in practice.

International AI standards

The EU AI Act heavily relies on European AI standardisation. European harmonisation bodies (CEN and CENELEC) are tasked with developing AI standards upon request from the European Commission. However, the delivery of their harmonised norms has been postponed from April to August this year, and doubts persist about whether the new deadline will be respected. While European AI standards are still in development, and the U.S. continues to resist comprehensive AI regulation, the ISO/IEC Joint Technical Committee on AI (SC 42) has been working on international AI standards since 2017 (see for example ISO/IEC 42001).

Interestingly, some countries have long sought to influence this regulatory forum by shaping international standards for AI. China, for instance, has focused intensively on international technical standards from a very early stage. Since 2018, it has connected the goal of rapid AI development with the ambition to shape international standards – positioning itself as a first mover in technology. It has thus interpreted its regulatory power as a form of “standard‑setting power”. Japan is also actively engaged at the international level: in 2023, it launched the G7 Hiroshima Process, a platform that brings together 55 countries to foster global AI cooperation. This complements its active involvement in ISO and IEC. The U.S., however, is far from absent: ANSI, the American standards organisation, hosts the ISO/IEC Joint Technical Committee on AI, and the committee is chaired by a U.S. representative.

When it comes to AI standards, the EU does not appear to be setting its characteristic Brussels effect. Certainly, the EU aims to shield its legal environment from external influence by strengthening the global role of the European standardisation system – pursuing a more strategic approach to international standardisation and fostering better coordination among EU Member States, standardisation bodies, and industries. At the same time, however, it strives to advance its commitment to fundamental rights and democratic values through international alliances in standard‑setting. With regard to AI, it would not be surprising if the EU were to adopt standards similar to those at the international level, as agreements between international and European standardisation organisations call for alignment of their standards (Vienna Agreement, Frankfurt Agreement) – an obligation which resonates in Recitals 3 and 6 of the European Regulation on Standardisation (Regulation 1025/2012).

Legal integration of international AI standards

As AI standards lack the status of formal law, their legal relevance depends on integration into the respective legal systems. This is where the European model of AI governance is particularly strong: Under Article 40 of the EU AI Act, if an AI system follows European harmonised AI standards, it is presumed to comply with the legal requirements of the AI Act. Are countries such as Brazil, with a history of aligning with European regulatory models, adopting this approach to integrating AI standards into law through a presumption of conformity?

Brazil is currently discussing Projeto de Lei 2338/2023 in the Federal Senate. This comprehensive act, like the EU AI Act, is built on a tiered framework of risk categories, ranging from unacceptable to minimal risk. In the highest risk category, certain AI systems are banned in Brazil, as they are in the EU. Similarly, the Brazilian AI Bill proposal places high-risk AI systems at the centre of its regulatory focus. This might seem like a perfect description of the Brussels effect (de iure), particularly given the close involvement of European experts in the legislative process. However, from the outset, the proposed AI Bill has, for example, placed a different emphasis on biometric remote identification, allowing for broader exceptions (Article 13). What is particularly noteworthy is that Brazil’s AI regulation acknowledges the importance of international standards; however, the responsibility for integrating them into the domestic legal framework is assigned to the SIA – the federal AI authority, set to operate as the National Data Protection Authority – without further specifying the process or actors involved. Legally, the Brazilian AI Bill proposal thus remains somewhat vague and deviates from the EU’s ‘presumption of conformity’ approach.

The Brussels mirage

While the EU AI Act continues to serve as a reference point for countries developing AI legislation, it is gradually turning more into a symbolic model than a practical blueprint. At closer look, there is thus no Brussels effect in AI regulation. Instead, the EU AI Act’s more nuanced international influence might be characterised as a Brussels mirage.

This image serves to highlight three key conclusions: First, a mirage is an optical illusion that shifts with perspective – what seems to be in one place is, on closer inspection, found to be somewhere else. In this sense, the EU AI Act’s regulatory approach is not apt for a Brussels effect. Its reliance on (international) AI standards makes it far more receptive to international developments than one might expect. However, in the realm of technical standard-setting, other countries seem to be outpacing the EU. Whether the EU will be able to meaningfully address the socio-technical dimension of AI standards at the international stage remains an open question.

Second, countries like Brazil are increasingly adding their own distinct emphasis to AI regulation, which in turn makes the contours of the EU AI Act’s influence beyond Europe appear less defined. The proposed AI Bill has undergone – and continues to undergo – substantial changes throughout the legislative process, partly as a result of growing public consultation. In light of the Brazilian Constitution (see Article 225), it confers individual and collective rights (such as the right to information and the right to obtain explanation of and to contest AI-based decisions). Relatedly, the Brazilian AI Bill proposal puts a different focus on societal risks, mentioning as its objective the protection of workers’ rights and the prevention of discrimination. Moreover, the Brazilian proposed AI Bill outlines the obligations of public actors more precisely than the EU AI Act. This reflects the fact that many public bodies in Brazil, including the judiciary, are using AI technologies to fulfil their tasks.

Finally, the picture of the Brussels mirage underscores that numerous questions remain open regarding AI regulation. For instance, the latest developments in the Brazilian legislative process indicate an abandonment of the current proposal towards a more principle-based regulation. Thus, the Brussels shift away from the ‘regulatory more and more’ approach appears to be echoed across the Atlantic and the Brussels effect seems to vanish into a mirage.